Sentiment Analysis of Netflix App Reviews on Google Play Store using the Naive Bayes

Abstract

The rapid growth of movie streaming apps like Netflix in Indonesia has generated millions of user reviews on the Google Play Store. However, the high volume of reviews makes it difficult for developers to understand user perceptions. This study aims to analyze sentiment towards 1,000 user reviews of the Netflix application collected through data scraping in the period August 16 to October 7, 2025. Labeling was conducted using a lexicon-based approach, where each word was assigned a score based on positive and negative dictionaries. The total score determined the sentiment polarity; if the score was greater than -1 (if score > -1), the review was categorized as positive, otherwise as negative. This approach helps reduce bias that may occur when labeling relies solely on user ratings.  The method used is Naïve Bayes with preprocessing stages including case folding, data cleaning, word normalization, tokenizing, stopword removal, and stemming. Furthermore, the data is weighted using TF-IDF and divided into training and test data with a ratio of 80:20. The implementation results show high model performance with accuracy of 77.96%, precision of 77.97%, recall of 77.96%, and an F1-score of 76.49%. The sentiment classification indicated that 36.5% of the reviews were positive and 63.5% were negative, indicating dissatisfaction of the majority of users. This study proves that the Naïve Bayes method is quite effective in classifying the sentiment of reviews. For further research, it is recommended to use a larger amount of data, a longer period, and explore other algorithms such as SVM, Random Forest, K-Nearest Neighbor, Decision Tree, and Rule-Based to improve the quality of sentiment analysis.